878 Haco OH H3CO OI HN Y H3C BU46 Bermejo Fm. et aL. 1999 irundt P. et al. 2003 Hac O BU74 KT-95 TRK-820 CoMFA) and comparative molecular similarity indices an interval of 15% on rotatable bonds to obtain their lowest analysis( CoMSIA) energy conformations. Finally, all molecules were mini- mized using the Tripos force field [16]. The energy. minimized conformation of molecule 17. which showed Materials and methods the highest binding affinity to the k-receptor of all compounds studied in this paper, served as the template Data set A well-known opiate analgesic pharmacophore-a tyra mine fragment [17] was common among all molecules Thirty-three compounds were collected from several Ihis fragment was used as the common substructure for structural alignments. The alignment resulting from the reports by the Lewis and Husbands group(see Table 1) alignment facility in SYBYL is shown in Fig. 2 [7-9, 11-13. The binding affinities to the K-opioid recep tor were determined by displacement binding assays of cloned human opioid receptors transferred onto CHO Cells PLS anal Seven of 33 compounds from the dataset were randor selected to form the test set, with the remaining 26 com- The Ki value to the K-receptor of each molecule was ds used to make the training set. converted to pK(logKi) and set as dependent values while CoMFA and CoMSIA descriptors were set as independent variables to perform PLS regression analyses Molecular modeling and structural alignment The CoMfa cutoff values were set to 30 kcal mol for both steric and electrostatic fields and all fields were scaled All calculations were carried out on a R14000 SGI Fuel by the default options in SYBYL workstation running the molecular modeling software The initial predictive coefficient q2 values and the 9 The initial structures of the 3 3 molecules were built (leave-one-out) cross-validation method. The cross-vali- ckage SYBYL V6.9[14] optimal number of components were obtained by the Loo ased on crystal structures of their analogs etorphine, dated coefficient q was calculated using the following diprenorphine and oxymorphone[15]. All molecules were formula: set in their unprotonated states and assigned Gasteiger- Huckel charges available in SYBYL. Except for the rigid morphine-like skeleton, random searches were performed ∑(prat-7 actual) on additional ring systems to ensure that their conforma- 9=1.0 tions were energetically favorable. For molecules with more flexibility, systematic searches were carried out with(CoMFA) and comparative molecular similarity indices analysis (CoMSIA). Materials and methods Data set Thirty-three compounds were collected from several reports by the Lewis and Husbands group (see Table 1) [7–9, 11–13]. The binding affinities to the κ-opioid recep￾tor were determined by displacement binding assays of cloned human opioid receptors transferred onto CHO Cells with [3 H]U69593 as the label. Seven of 33 compounds from the dataset were randomly selected to form the test set, with the remaining 26 com￾pounds used to make the training set. Molecular modeling and structural alignment All calculations were carried out on a R14000 SGI Fuel workstation running the molecular modeling software package SYBYL v6.9 [14]. The initial structures of the 33 molecules were built based on crystal structures of their analogs etorphine, diprenorphine and oxymorphone [15]. All molecules were set in their unprotonated states and assigned Gasteiger– Hückel charges available in SYBYL. Except for the rigid morphine-like skeleton, random searches were performed on additional ring systems to ensure that their conforma￾tions were energetically favorable. For molecules with more flexibility, systematic searches were carried out with an interval of 15° on rotatable bonds to obtain their lowest energy conformations. Finally, all molecules were mini￾mized using the Tripos force field [16]. The energy￾minimized conformation of molecule 17, which showed the highest binding affinity to the κ-receptor of all compounds studied in this paper, served as the template. A well-known opiate analgesic pharmacophore—a tyra￾mine fragment [17] was common among all molecules. This fragment was used as the common substructure for structural alignments. The alignment resulting from the alignment facility in SYBYL is shown in Fig. 2. PLS analysis The Ki value to the κ-receptor of each molecule was converted to pKi(−logKi) and set as dependent values, while CoMFA and CoMSIA descriptors were set as independent variables to perform PLS regression analyses. The CoMFA cutoff values were set to 30 kcal mol−1 for both steric and electrostatic fields and all fields were scaled by the default options in SYBYL. The initial predictive coefficient q 2 values and the optimal number of components were obtained by the LOO (leave-one-out) cross-validation method. The cross-vali￾dated coefficient q 2 was calculated using the following formula: q2 ¼ 1:0
P γ ðγpred
γactualÞ 2 P γ γactual
γmean ð Þ2 N HO O H3CO OH H 7 8 BU46 N H3CO OH O CH3 Ph H3C 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 10 19 Bermejo FM, et al. 1999 N H3CO OH CH3 Ph HN H3C Grundt P, et al. 2003 N HO H3CO CH3 HN Ph O AcO N S O CH3 O N OH HO O N H3C O O BU74 KT-95 TRK-820 Fig. 1 Some opioids with potent κ-agonist potency 878